- Email: [email protected]
Thresholds in the relationship between mortality and left ventricular hypertrophy defined by electrocardiography
Available online at www.sciencedirect.com
Journal of Electrocardiology 41 (2008) 342 – 350 www.jecgonline.com
Thresholds in the relationship between mortality and left ventricular hypertrophy defined by electrocardiography☆ Edward P. Havranek, MD, a,b, * Caroline D.B. Emsermann, MS, b Desiree N. Froshaug, BS, b Frederick A. Masoudi, MD MSPH, a,b Mori J. Krantz, MD, a,b Rebecca Hanratty, MD, a,b Raymond O. Estacio, MD, a,b L. Miriam Dickinson, PhD, b John F. Steiner, MD, MPH b b
a Department of Medicine, Denver Health Medical Center, Denver, Colorado, USA Colorado Health Outcomes Program, University of Colorado at Denver and Health Sciences Center, Denver, Colorado, USA Received 13 February 2007; accepted 15 November 2007
Abstract
Background: Electrocardiographic criteria for the diagnosis of left ventricular hypertrophy in current use were defined using autopsy results or echocardiography; criteria defined using mortality might be more clinically meaningful. Methods: Using data from Third National Health and Nutrition Examination Survey (NHANES III), we selected electrocardiographic measures that best differentiated those surviving at 5 years from those who did not. We identified voltage thresholds using regression techniques and then compared survival for subjects above and below the thresholds. Results: Cornell voltage, Cornell product, and Novacode estimate of left ventricular mass index were discriminative for mortality and had identifiable thresholds present in their relationships with mortality. Independent of systolic blood pressure, there were significant associations with 5-year mortality for Novacode index above threshold; hazard ratios were 1.58 for women and 1.27 for men, and for 5-year cardiovascular mortality were 1.78 for women and 2.34 for men. Conclusions: Electrocardiographic criteria for left ventricular hypertrophy validated against mortality might be clinically useful. © 2008 Elsevier Inc. All rights reserved.
Keywords:
Left ventricular hypertrophy; Electrocardiography; Mortality; Survival; NHANES
Introduction Left ventricular hypertrophy (LVH) is a significant risk factor for cardiovascular disease independent of blood pressure,1,2 and reductions in left ventricular mass are associated with lower cardiovascular event rates independent of reductions in blood pressure. In the Losartan Intervention for Endpoint reduction (LIFE) study, the risk of cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke was reduced by 22% for each one SD decrement in left ventricular mass index measured by echocardiography, independent of systolic blood pressure or treatment assigned.3 In the Heart Outcomes Prevention Evaluation (HOPE) trial, patients who had resolution of electrocardio☆ National Heart, Lung, and Blood Institute grant U01 HL079160 (Bethesda, Maryland, USA) and Health Resources and Services Administration Administrative Unit grant HP00054 5 D12 (Rockville, Maryland, USA). ⁎ Corresponding author. Denver Health Medical Center #0960, Denver, CO 80204-4507, USA. Tel.: +1 303 436 5499; fax: +1 303 436 7739. E-mail address: [email protected]
0022-0736/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.jelectrocard.2007.11.006
graphic LVH findings or who did not develop new LVH findings had significantly fewer cardiovascular deaths than did patients who did not have resolution or who developed new LVH.4 In the Appropriate Blood Pressure Control in Diabetes (ABCD) trial, the change in a QRS voltage measure of LVH predicted time to cardiovascular events, after controlling for change in blood pressure.5 These findings imply that hypertension treatment that leads to both regression of LVH and blood pressure reduction to goal may decrease cardiovascular event rates more than treatment to a blood pressure goal alone. To be clinically meaningful, a definition of LVH should be linked to a mortality reduction and make use of readily available technology. The most readily available technologies used to define LVH in research and in clinical practice are echocardiography and electrocardiography. Echocardiography has better sensitivity and specificity for detection of LVH than electrocardiography but is expensive and is technically inadequate in 10% to 20% of unselected subjects1 and in more than 30% in of older subjects.6 In contrast, electrocardiography is inexpensive,
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
is readily available, and can be performed in nearly all patients and interpreted with relatively little advanced skill. We hypothesized that electrocardiographic voltage measures of LVH treated as continuous variables have relationships with mortality that can be modeled to define a threshold above which mortality rises progressively, independent of blood pressure. Methods Data sources The data sources for the study were the Third National Health and Nutrition Examination Survey (NHANES III)7 database and the NHANES III linked mortality file.8 Third National Health and Nutrition Examination Survey was conducted in 81 counties in the United States between 1988 and 1994; subjects were selected to obtain a sample representative of the civilian, noninstitutionalized population of the United States greater than 2 months old. Electrocardiograms were obtained in NHANES III only for subjects older than 40 years. Twelve-lead tracings were recorded during examination at a mobile examination center by trained technicians using a Marquette MAC 12 system (GE Healthcare, UK). Electrocardiograms were transferred to the National Center for Health Statistics (Hyattsville, MD) for electronic analysis. Baseline blood pressure was measured as the average of the 3 recordings performed at the mobile examination center by a physician. Examiners used a mercury manometer and were instructed to record the blood pressure from the right arm with the subject seated, choosing the appropriate cuff size from among four available. The NHANES III linked mortality file contains the results of linking NHANES III subject identifiers with data in the National Death Index9 as of December 31, 2000. Date of death and cause of death available in the National Death Index are from death certificates. Definitions We chose to examine electrocardiographic measures of LVH that had been validated by echocardiography and that could be analyzed as continuous variables. Using these criteria we selected 6 measures. The Cornell voltage10 is the sum of the R wave dimension in lead aVL and the S wave dimension in lead V3; LVH is said to be present if this voltage is greater than 20 mm in women and 28 mm in men. The Cornell product11 is the product of the Cornell voltage and the QRS duration; LVH is said to be present if this product is greater than 2436 mm-ms (adjusted by the addition of 8 mm for female sex). The adjusted Cornell voltage is defined by the regression equations of Norman and Levy12: For males: ðCornell voltageþ0:0174 * ageþ0:1914 * BMI4:0Þ=2 For females: ðCornell voltageþ0:0387 * ageþ0:1778 * BMI4:9Þ=3 where BMI is the body mass index.
343 13
The Sokolow-Lyon voltage has been defined variously in the literature; we chose to define it as the sum of the greatest S wave dimension in leads V1 or V2 and the greatest R wave dimension in leads V5 or V6; LVH is said to be present if this voltage is greater than 35 mm. Finally, we used the Novacode estimates of left ventricular mass and left ventricular mass index calculated from the regression equations of Rautaharju et al.14 The variables and coefficients for these equations are available in the NHANES III electrocardiography data file documentation15 and in the Appendix. We also examined the independent effect of the ST-T changes associated with LVH (the so-called “strain pattern”) on outcomes, both separately and in combination with Cornell voltage and Cornell product. A strain pattern was judged to be present if there was horizontal or downsloping ST depression associated with T wave inversion in lead V5 and/or V6.16 We defined an analysis set by including those subjects with all the necessary data available to calculate the electrocardiographic measures and whose vital status was known. We excluded subjects with QRS duration greater than 0.12 seconds because validity in the presence of bundle branch block has not been established for all of the electrocardiographic measures. We defined 5-year mortality as the primary outcome of interest. All-cause mortality is less subject to ascertainment bias than is cardiovascular mortality, and the 5-year time frame has been commonly used in cardiovascular epidemiology as a compromise between the longer times necessary for effects of the exposure (in this case LVH) to accrue and the shorter times that exclude the accrual of competing causes of mortality. We defined cardiovascular death as those for which the principal cause of death was listed on the death certificate as International Classification of Diseases, Ninth Revision (ICD-9) codes 390.0-459.9 (for deaths before or during 1999) or ICD, 10th Revision (ICD-10) codes I10-15, I20-25, I50-51, I60-69, I70-73 (for deaths after 1999). In assessing the independent contribution of electrocardiographic LVH measures to mortality risk, we defined the relevant covariates as the Framingham risk factors17 of age, sex, smoking status, presence or absence of diabetes, systolic blood pressure, and total and high-density lipoprotein cholesterol. Diabetes was defined in NHANES III by self-identification. Statistical analysis We assessed for colinearity among the 6 measures using Pearson correlation coefficients. We selected the electrocardiographic measures best able to discriminate those at risk for mortality from those not at risk using receiver operating characteristic (ROC) curves. We divided the sample into deciles (with an equal number of subjects in each group), plotted 5-year all-cause mortality by decile, and calculated sensitivity and specificity for 5-year mortality by decile. We constructed a separate ROC curve for each of the 6 measures and calculated the area under the ROC curve. Measures were considered discriminative if the
344
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
lower bound of the 95% confidence intervals around the areas under the curve was greater than 0.50. For the discriminative measures, we sought thresholds above which mortality rose above a baseline level. Because left ventricular mass and left ventricular mass index vary systematically by sex, we sought separate thresholds for men and women. We selected at random two thirds of the analysis sample for a development sample. To enhance interpretability, we recoded the data for this set by 10-unit ranges for the measure, except for the smallest and largest of the ranges that included the remaining values on either end, for a total of 10 ranges. We applied piecewise regression to this set with 5-year mortality as the outcome, modeling the data as a spline function with 2 distinct slopes. We took as our thresholds the midpoint of the ranges for which risk was significantly greater than for values below that category. We tested the performance of the thresholds we found in 4 ways: First, we compared 5-year survival curves for male and female subjects with baseline voltage measures above and below the thresholds using the log-rank test. Second, we used Cox proportional hazards models to assess for an independent contribution of voltage, first controlling for systolic blood pressure alone and then controlling for all the Framingham risk factors, to the primary outcome of 5-year all-cause mortality. Third, we created the same Cox proportional hazards models for 3 secondary outcomes: 5-year cardiovascular mortality, 10-year all-cause mortality, and 10-year cardiovascular mortality. Finally, we repeated the Cox models in the one third of the sample that was not used to establish the thresholds. Results Of subjects in the NHANES III sample, 8561 had an electrocardiogram performed. We excluded from our
analysis sample the 641 subjects with QRS duration more than 0.12 second, 4 subjects because of incomplete electrocardiographic data, and 4 subjects because vital status was not known, leaving 7852 subjects with data available for analysis. In addition, 379 subjects were excluded from the adjusted analyses in the validation set because of missing Framingham risk factors. The characteristics of the study sample are shown in Table 1 and reflect a middle-aged and elderly population. The mean age of the sample was approximately 60 years and was slightly female predominant. Approximately 10% of subjects had diabetes, and approximately 40% had systolic hypertension. Correlation coefficients among the voltage measures are shown in Table 2; with the possible exception of the Cornell voltage and Cornell product, the measures are sufficiently different from one another to justify examination of each of them. Fig. 1 shows 5-year all-cause mortality by decile for each of the 6 electrocardiographic measures, and Table 3 shows the area under the ROC curve (C-statistic) for each of the 6 measures. Four measures, Cornell voltage, Cornell product, adjusted Cornell voltage, and Novacode estimate of left ventricular mass index, were discriminative for mortality, with C-statistics significantly greater than 0.5. Because we examined LVH as a univariate predictor of mortality, the C-statistics are much lower than that typically described for multivariable models. By piecewise regression analysis, there were statistically identifiable thresholds in the mortality relationship for 3 of the discriminative measures: Cornell voltage, Cornell product, and Novacode left ventricular mass index; these thresholds are shown in Table 3. In men, 16.7% of the sample had a Novacode left ventricular mass index above threshold, 16.1% had a Cornell voltage above threshold, and 13.3% had Cornell product above threshold. Comparable figures for women were 19.3%, 9.2%, and 13.2%, respectively.
Table 1 Characteristics of study population Variable
Total
Males
Females
Age (y, mean [SD]) Sex (% female) Current smoker at baseline (%) Total cholesterol (mmol/L, mean [SD]) HDL cholesterol (mmol/L, mean [SD]) Diabetes (%) Systolic BP (mm Hg) Cornell voltage (mm, mean [SD]; median, IQR) Cornell product, mm-ms, (mean [SD]; median, IQR) Adjusted Cornell voltage (mean [SD]; median [IQR]) Sokolow-Lyons voltage (mm, mean [SD]; median [IQR]) Novacode mass (g, mean [SD]; median [IQR]) Novacode mass index (g/m2, mean [SD]; median [IQR])
59.9 (13.4) 53 23 5.64 (1.14) 1.32 (0.42) 11 132.9 (19.9) 13.1 (5.5); 12.7 (9.20-16.5)
60 (13.4) 27 5.51 (1.10) 1.19 (0.38) 11 134.0 (18.5) 14.0 (5.7); 14.0 (10.3-17.9)
59.7 (13.5) 18 5.78 (1.16) 1.42 (0.42) 11 132.4 (21.0) 12.0 (5.0); 11.6 (8.38-15.1)
1272 (580); 1205 (858-1611)
1428 (610); 1368 (999-1788)
1132 (512); 1076 (761-1432)
1.5 (0.6); 1.5 (1.08-1.90)
1.8 (0.6); 1.8 (1.44-2.19)
1.2 (0.5); 1.2 (0.88-1.50)
25.9 (7.8); 25.0 (20.6-30.2)
27 (8.3); 26.3 (21.5-32.1)
24.8 (7.0); 23.8 (20.0-28.7)
154 (29.4); 153 (133-174)
172 (25.5); 171 (156-188)
138 (22.6); 136 (123-152)
105 (21.7); 103 (90-118)
112 (20.5); 110 (99-123)
98.4 (20.8); 96.0 (85-110)
HDL indicates high-density lipoprotein; BP, blood pressure; IQR, interquartile range.
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
345
Table 2 Pearson correlation coefficient matrix
Cornell voltage Cornell product Adjusted Cornell Sokolow-Lyons Novacode LVM Novacode LVMI
Cornell voltage
Cornell product
Adjusted Cornell
Sokolow-Lyons
Novacode LVM
Novacode LVMI
1.0 0.942 0.628 0.397 0.566 0.465
1.0 0.621 0.357 0.631 0.585
1.0 0.093 0.855 0.393
1.0 0.095 0.401
1.0 0.471
1.0
LVM indicates left ventricular mass; LVMI, left ventricular mass index.
As described in the Methods section, we tested the thresholds in 4 ways. First, we compared 5-year survival for women and for men by voltage above and below the thresholds. As shown in Figs. 2, 3, and 4, there are statistically significant differences for all 3 sets of thresholds. Second, we assessed for an independent contribution of the voltages to 5-year mortality by adjusting first for systolic blood pressure alone and then for all the Framingham risk factors. As shown in Table 4, statistically significant relationships for men are present with adjustment for all the Framingham risk factors but not for blood pressure alone. For women, there are statistically significant relationships
adjusted for blood pressure but not for all the Framingham risk factors. Third, we created the same Cox proportional hazards models for the 3 secondary outcome measures of 5-year cardiovascular mortality, 10-year all-cause mortality, and 10-year cardiovascular mortality. Adjusted relationships were stronger for cardiovascular mortality than for allcause mortality. Fig. 5 shows the results for the adjustment for systolic blood pressure alone. The Novacode left ventricular mass index predicted mortality independent of blood pressure for all 3 secondary outcome measures; results with the other 2 measures were less consistent. With adjustment for all Framingham risk factors in men, the
Fig. 1. Five-year all-cause mortality by decile for each of the voltage criteria tested.
346
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
Table 3 C-statistics and sex-specific thresholds for each of the voltage criteria Voltage measure
C-statistic (95% confidence interval)
Threshold—females
Threshold—males
Cornell voltage Cornell product Adjusted Cornell voltage Sokolow-Lyons voltage Novacode LVM Novacode LVMI
0.55 (0.53-0.57) 0.55 (0.56-0.57) 0.56 (0.54-0.59) 0.52 (0.49-0.54) 0.50 (0.48-0.52) 0.59 (0.57-0.61)
19 mm 1700 mm-ms Not applicable ⁎ Not applicable Not applicable 115 g/m2
20 mm 2150 mm-ms Not applicable Not applicable Not applicable 130 g/m2
⁎ Piecewise regression demonstrated there was no threshold for the adjusted Cornell voltage.
thresholds for all 3 measures predicted mortality independently for all 3 secondary outcomes. For women, the thresholds predicted mortality independent of all Framingham risk factors only for Novacode left ventricular mass index and 10-year cardiovascular mortality. Finally, we tested these thresholds in the one third of the sample not used to develop the thresholds. As shown in Table 5, hazard ratios were consistent with those found in the two thirds of the sample used to develop the thresholds for Novacode left ventricular mass index but not for the Cornell voltage and Cornell product. Results were similar for the secondary outcomes in the validation set. We performed additional analysis to assess whether the ST-T changes accounted for in the Novacode left ventricular mass index were responsible for the superior performance of this measure. First, we assessed the independent contribution of the strain pattern alone. In males in the development set, the hazard ratio associated with strain for 5-year all-cause
Fig. 2. A, Kaplan-Meier curves comparing survival for men by Novacode left ventricular mass index above and below threshold. B, Kaplan-Meier curves comparing survival for women by Novacode left ventricular mass index above and below threshold.
mortality was 5.72 (3.67-8.92) unadjusted, 4.56 (2.88-7.23) adjusted for systolic blood pressure, and 2.56 (1.59-4.13) adjusted for all Framingham risk factors; comparable hazard ratios for women were 4.84 (2.28-10.29), 2.44 (1.13-5.29), and 2.11 (0.93-4.81), respectively. Results in the validation set confirmed these findings. The value of this strong association between the strain pattern and subsequent mortality was tempered by its relatively low prevalence. A strain pattern was found on the electrocardiograms of only 5.0% of the 760 subjects who had died at 5 years, whereas 18.5% of subjects with Cornell voltage above threshold, 18.2% of subjects with Cornell product above threshold, and 26.8% of subjects with Novacode left ventricular mass index above threshold had died at 5 years. We therefore tested association between presence of a strain pattern in combination with Cornell voltage and Cornell product. In males in the development set, the hazard ratio associated with strain combined with Cornell voltage for 5-year all-cause mortality was 1.68 (1.30-2.18) unadjusted, 1.42 (1.09-1.86) adjusted for systolic blood pressure,
Fig. 3. A, Kaplan-Meier curves comparing survival for men by Cornell voltage above and below threshold. B, Kaplan-Meier curves comparing survival for women by Cornell voltage above and below threshold.
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
347
These results suggest that electrocardiography might be used to establish regression of left ventricular hypertrophy in the management of hypertension, which in turn might be expected to reduce mortality. Even small increases in voltages above the thresholds appear to be associated with significant increases in mortality. Comparison with prior literature
Fig. 4. A, Kaplan-Meier curves comparing survival for men by Cornell product above and below threshold. B, Kaplan-Meier curves comparing survival for women by Cornell product above and below threshold.
and 1.47 (1.12-1.94) adjusted for all Framingham risk factors; comparable hazard ratios for women were 2.85 (2.04-3.97), 1.76 (1.23-2.52), and 1.44 (0.99-2.10), respectively. These results were marginally better than the Novacode measure in men and marginally worse than the Novacode measure in women and were nearly identical for the Cornell product. Results in the validation set confirmed these findings. Of the 760 subjects who died, 21.4% had strain or Cornell voltage criteria present and 21.2% had strain or Cornell product present. Discussion Our study confirmed the relationship between greater electrocardiographic voltage measures of LVH and higher mortality and identified the Cornell voltage, Cornell product, and Novacode estimate of left ventricular mass index as having discriminative capacity for mortality and having demonstrable thresholds above which mortality rises progressively. The Novacode left ventricular mass index performed slightly better in multivariable models. The Cornell voltage and Cornell product also demonstrated significant threshold relationships, but the thresholds showed fewer significant relationships with mortality after adjustment for other risk factors. Novacode left ventricular mass index above threshold was present in 26.8% of subjects who had died at 5 years; for Cornell voltage and Cornell product these figures were 18.5% and 18.2%, respectively. There were fewer statistically significant relationships in a validation sample comprising one third of the full analytic sample probably because of the smaller number of observations in the validation sample.
Electrocardiographic measures of LVH have been associated with elevated risk of mortality in a number of prior studies, though few have treated these measures as continuous variable. Prineas and colleagues18 showed the Sokolow-Lyons voltage, Cornell voltage, Cornell product, and Novacode score used as continuous variables were better predictors of mortality in patients in the Multiple Risk Factor Intervention Trial than were the same parameters dichotomized as the presence or absence of LVH. The electrocardiographic measure that performed best in our study, the Novacode estimate of left ventricular mass index, is probably the least studied of the measures we investigated. In one study,19 this score had lower sensitivity and specificity for the detection of echocardiographic LVH than the Sokolow-Lyons voltage, the Cornell voltage, and the Cornell product. In a study of subjects in the Cardiovascular Health Study,6 the Novacode equations consistently overestimated left ventricular mass. The superior sensitivity and specificity for detection of mortality despite its limitations for detection of echocardiographic hypertrophy may be a result of the unique use of T-wave inversion in the Novacode score. Inversion of the T wave is a feature of the “strain” pattern that is an independent predictor of poor prognosis.20 This explanation is supported by the strong association between the presence of a strain pattern and subsequent mortality in our data set and by combination of the strain pattern with either Cornell voltage or Cornell product made the performance of these 2 measures comparable to that of the Novacode index. Further Table 4 Associations between 5-year total mortality and electrocardiographic indices above threshold for the development set Males (309 deaths)
Unadjusted Adjusted for systolic BP Adjusted for all Framingham risk factors ⁎
Cornell voltage
Cornell product
Novacode LVMI
HR
HR
HR
95% CI
95% CI
95% CI
1.42 1.08-1.87 1.38 1.03-1.86 1.52 1.16-1.98 1.19 0.89-1.58 1.16 0.85-1.57 1.27 0.99-1.68 1.34 1.01-1.79 1.34 0.98-1.82 1.40 1.05-1.85
Females (201 deaths) Unadjusted 2.66 1.89-3.75 2.33 1.70-3.20 2.33 1.74-3.12 Adjusted for 1.65 1.15-2.24 1.60 1.15-2.24 1.58 1.16-2.16 systolic BP Adjusted for 1.34 0.91-1.98 1.41 0.99-2.01 1.12 0.81-1.54 all Framingham risk factors HR indicates hazard ratio; CI, confidence interval. ⁎ Age, smoking status, diabetes, systolic blood pressure, total cholesterol, HDL cholesterol.
348
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
Fig. 5. Hazard ratios with 95% confidence intervals for the secondary outcome measures, adjusted for systolic blood pressure, for men (top panel) and for women (bottom panel).
development and validation of such combinations are worthy of future study. The thresholds we identified for the Novacode left ventricular mass index are strikingly comparable to the upper limits of normal for echocardiographic left ventricular mass index determined by Devereux and colleagues21 (132136 g/m2 for men and 109-112 g/m2 for women). In an article using data from the Framingham study and the same echocardiographic conventions as Devereux et al, the thresholds identified were similar for men (131 g/m2) and slightly lower for women (100 g/m2).22 The hazard ratios for mortality in the current study are similar to the results of studies based on echocardiography for men but not for women. In a report from the Framingham study, Levy et al1 found relative risk of 1.49 in men for allcause mortality adjusted for age and risk factors and 1.73 for cardiovascular mortality; for women, the comparable figures were 2.01 and 2.12, respectively. Liao and colleagues23 studied the relationship between echocardiographically
determined left ventricular mass and subsequent mortality and, after adjusting for age and hypertension, also found a stronger relationship between LVH and mortality in women compared to men. In the current study, the opposite is the case, with relationships stronger in men than in women. This suggests that further refinement of the Novacode equations for women is possible and might further improve the discriminative capacity of the thresholds. Although the Novacode left ventricular mass index threshold was more clearly predictive of mortality than the Cornell voltage or the Cornell product thresholds, both of these latter measures were discriminative for mortality and had identifiable thresholds. For men, we found an increase in mortality at lower levels than the established criteria for diagnosis of LVH for both measures (20 vs 28 mm for the Cornell voltage and 2150 vs 2436 mm-ms for the Cornell product). For women, we found thresholds at comparable levels (19 vs 20 mm for the Cornell voltage and 1700 vs approximately 1740 mm-ms for the Cornell product). The
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350 Table 5 Associations between 5-year total mortality and electrocardiographic indices above threshold for the validation set Males (153 deaths)
Unadjusted Adjusted for systolic BP Adjusted for all Framingham risk factors ⁎
Cornell voltage
Cornell product
Novacode LVMI
HR
HR
95% CI
HR
1.02 0.64-1.62 1.08 0.82 0.51-1.31 0.88
0.65-1.80 0.53-1.46
1.43 0.97-2.11 1.14 0.76-1.70
0.86 0.53-1.40 0.91
0.53-1.55
1.29 0.86-1.92
95% CI
Females (118 deaths) Unadjusted 1.83 1.04-3.21 1.45 Adjusted for 1.47 0.82-2.64 1.18 systolic BP Adjusted for 1.03 0.53-1.97 0.84 all Framingham risk factors
95% CI
349
attenuated when the higher prevalence of obesity in African Americans is accounted for.35,36 The Novacode estimate of left ventricular mass index not only accounts for differences in body size but also uses separate regression equations for black and white women. Other potential ethnic differences, particularly for Latinos, have generally not been explored in the literature. The relationship among race and ethnicity, electrocardiographic measures of LVH, and mortality is deserving of future research. Conclusions and implications
0.87-2.42 0.70-2.00
2.19 1.44-3.32 1.84 1.17-2.88
0.48-1.49
1.21 0.76-1.92
⁎ Age, smoking status, diabetes, systolic blood pressure, total cholesterol, HDL cholesterol.
Electrocardiographic definitions of LVH based on relationships with mortality are feasible. Such definitions might be clinically useful in the routine management of patients with hypertension as goals for therapy in addition to lowering blood pressure below goal; independent prospective validation is required. Appendix Men:
reasons for this sex difference in the relationship between electrocardiographic voltage and outcome are uncertain. One possibility is that the partition values for Cornell voltage and Cornell product were set too high and that partitions with lower sensitivities would be more clinically useful. The Cornell voltage and the Cornell product are simple to calculate, might therefore be more useful in routine clinical practice, and are worthy of further study. The poor performance of the Sokolow-Lyons voltage is in part because of its low sensitivity and specificity for the detection of LVH,24-28 but other factors may be at play. A study by Velury and Spodick29 suggested that low R voltage (b10 mm) in leads V5 and V6 has a high sensitivity for the diagnosis of LVH. This may explain the trend toward higher mortality in patients with low Sokolow-Lyons voltage seen in Fig. 1 because low Sokolow-Lyon voltage might also be a marker of LVH. Study limitations The greatest limitation to the current study is that it relies on data taken at a single point in time. We are thus unable to assess whether decrements in the electrocardiographically estimated left ventricular mass index are associated with diminished risk of mortality. Data from other studies,30-32 however, suggest an association between decrements in electrocardiographic voltage and diminished mortality. Information about potentially significant cofactors, such as treatment during follow-up, was not available. Another limitation is that we relied on cardiovascular causes of death determined from death certificates, which may be unreliable. However, we found significant relationships with all-cause mortality, an outcome less subject to ascertainment bias. We chose not to consider race or ethnicity as possible covariates. Although the prevalence of LVH assessed by Sokolow-Lyon or Cornell voltage criteria is greater in African Americans, there are no racial differences when assessed by echocardiography,33,34 and the diminished specificity of the electrocardiographic criteria are greatly
Variable R in amplitude in V5 Q or S amplitude in V1 Q or S amplitude in III Negative T amplitude in V6 Positive T amplitude in aVR QRS duration Intercept
Coefficient LV mass 0.0217 0.0338 0.0600 0.3158 −0.2958 1.8204 −58.5098
LV mass index 0.0100 0.0203 0.0287 0.1819 −0.1482 1.0485 −36.4290
Coefficient LV mass 0.0320 0.0233 0.0693 −0.1545 0.1122 – −0.1236 134.7722
LV mass index – 0.0178 0.0528 −0.1128 0.1075 0.1701 −0.0939 88.4357
White women:
Variable R in amplitude in aVL R amplitude in V5 Q or S amplitude in V5 Q or S amplitude in I Positive T amplitude in V1 Negative T amplitude in aVF Positive T amplitude in V6 Intercept Black women:
Variable R in amplitude in aVL R amplitude in I R amplitude V6 + S amplitude V2 R amplitude in V1 R amplitude in V2 Larger of Q or S amplitude in V6 Negative T amplitude in aVL Negative T amplitude in I QRS duration Intercept
Coefficient LV mass – 0.0498 0.0235 −0.0507 – −0.0980 – 0.5225 1.8478 −90.7136
LV mass index 0.0216 – 0.0184 – −0.0143 −0.0693 0.199 – 0.7460 −22.3064
References 1. Levy D, Garrison RJ, Savage DD, Kannel WB, Casatelli WP. Prognostic implications of echocardiographically determined left ventricular mass in the Framingham Heart Study. N Engl J Med 1990;322:1561.
350
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
2. Brown DW, Giles WH, Croft JB. Left ventricular hypertrophy as a predictor of coronary heart disease mortality and the effect of hypertension. Am Heart J 2000;140:848. 3. Devereux RB, Wachtell K, Gerdts E, et al. Prognostic significance of left ventricular mass change during treatment of hypertension. JAMA 2004;292:2350. 4. Matthew J, Sleight P, Lonn E, et al, for the Heart Outcomes Prevention Evaluation (HOPE) Investigators. Reduction of cardiovascular risk by regression of electrocardiographic markers of left ventricular hypertrophy by the angiotensin-converting enzyme inhibitor ramipril. Circulation 2001;104:1615. 5. Havranek EP, Esler A, Mehler PS, Estacio RO, Schrier RW. Differential effects of anti-hypertensive agents on ECG voltage: results from the Appropriate Blood Pressure Control in Diabetes (ABCD) Trial. Am Heart J 2003;145:993. 6. Rautaharju PM, Manolio TA, Siscovick D, et al, for the Cardiovascular Health Study Collaborative Research Group. Utility of new electrocardiographic models for left ventricular mass in older adults. Hypertension 1996;28:8. 7. http://www.cdc.gov/nchs/nhanes.htm. Accessed June 1, 2007. 8. http://www.cdc.gov/nchs/r&d/nchs_datalinkage/nhanes_data_linkage_ activities.htm. Accessed June 1, 2007. 9. http://www.cdc.gov/nchs/ndi.htm. Accessed June 1, 2007. 10. Casale PN, Devereux RB, Kligfield P, et al. Electrocardiographic detection of left ventricular hypertrophy: development and prospective validation of improved criteria. J Am Coll Cardiol 1985;6:572. 11. Molloy TJ, Okin PM, Devereux RB, Kligfield P. ECG detection of echocardiographic left ventricular hypertrophy by the simple QRS voltage-duration product. J Am Coll Cardiol 1992;20:1180. 12. Norman JE, Levy D. Improved electrocardiographic detection of echocardiographic left ventricular hypertrophy: results of a correlated data base approach. J Am Coll Cardiol 1995;26:1022. 13. Sokolow M, Lyon TP. The ventricular complex in left ventricular hypertrophy as obtained by unipolar precordial and limb leads. Am Heart J 1949;37:161. 14. Rautaharju PM, Lacroix AZ, Savage DD, et al. Electrocardiographic estimate of left ventricular mass versus radiographic cardiac size and the risk of cardiovascular disease mortality in the epidemiologic Follow-Up Study of the First National Health and Nutrition Examination Survey. Am J Cardiol 1988;62:59. 15. Available at http://www.cdc.gov/nchs/about/major/nhanes/nh3data. htm#Data%20Files%202a. Accessed June 1, 2007. 16. Okin PM, Devereux RB, Nieminen MS, et al, for the LIFE Study Investigators. Relationship of the electrocardiographic strain pattern to left ventricular structure and function in hypertensive patients: The LIFE Study. J Am Coll Cardiol 2001;38:514. 17. Anderson KM, Wilson PWF, Odell PM, Kannel WB. An updated coronary risk profile. A statement for health professionals. Circulation 1991;83:357. 18. Prineas RJ, Rautaharju PM, Grandits G, Crow R, for the MRFIT Research Group. Independent risk for cardiovascular disease predicted by modified continuous score ECG criteria for 6-year incidence and regression of left ventricular hypertrophy among clinically disease free men: 16-year follow-up for the Multiple Risk Factor Intervention Trial. J Electrocardiol 2001;34:91. 19. Okin PM, Roman MJ, Devereux RB, Kligfield P. Electrocardiographic identification of left ventricular hypertrophy: test performance in relation to definition of hypertrophy and presence of obesity. J Am Coll Cardiol 1996;27:124.
20. Verdecchia P, Schillaci G, Borgioni C, et al. Prognostic value of a new electrocardiographic method for diagnosis of left ventricular hypertrophy. J Am Coll Cardiol 1998;31:383. 21. Devereux RB, Lutas EM, Casale PN, et al. Standardization of M-mode echocardiographic left ventricular anatomic measurements. J Am Coll Cardiol 1984:1222. 22. Levy D, Savage DD, Garrison RJ, Anderson KM, Kannel WB, Castelli WP. Echocardiographic criteria for left ventricular hypertrophy: the Framingham Heart Study. Am J Cardiol 1987;59:956. 23. Liao Y, Cooper RS, Mensah GA, McGee DL. Left ventricular hypertrophy has a greater impact on survival in women than in men. Circulation 1995;92:805. 24. Casale PN, Devereux RB, Kligfield P, et al. ECG detection of left ventricular hypertrophy: development and prospective validation of improved criteria. J Am Coll Cardiol 1985;6:572. 25. Casale PN, Devereux RB, Alonso DR, Campo E, Kligfield P. Improved sex-specific criteria of left ventricular hypertrophy for clinical and computer interpretation of electrocardiograms: validation with autopsy findings. Circulation 1987;75:565. 26. Vijan SG, Manning G, Millar-Craig MW. How reliable is the electrocardiogram in detecting left ventricular hypertrophy in hypertension? Postgrad Med J 1991;67:646. 27. Abergel E, Tase M, Menard J, Chatellier G. Influence of obesity on the diagnostic value of ECG criteria for detecting left ventricular hypertrophy. Am J Cardiol 1996;77:739. 28. Okin PM, Jern S, Devereux RB, Kjeldsen SE, Dahlof B, for the LIFE Study Group. Effect of obesity on electrocardiographic left ventricular hypertrophy in hypertensive patients: The Losartan Intervention For Endpoint (LIFE) Reduction in Hypertension Study. Hypertension 2000;35:13. 29. Velury S, Spodick DH. Increased left ventricular mass in patients with lateral low-voltage electrocardiograms. Am J Cardiol 1992;69:707. 30. Levy D, Salomon M, D'Agostino RB, Belanger AJ, Kannel WB. Prognostic implications of baseline ECG features and their serial changes in subjects with left ventricular hypertrophy. Circulation 1994;90:1786. 31. Fagard RH, Staessen JA, Thijs L, et al, for the Systolic Hypertension in Europe (Syst-Eur) Trial Investigators. Prognostic significance of electrocardiographic voltages and their serial changes in elderly with systolic hypertension. Hypertension 2004;44:459. 32. Okin PM, Devereux RB, Jern S, et al, for the LIFE Study Investigators. Regression of electrocardiographic left ventricular hypertrophy during antihypertensive treatment and the prediction of major cardiovascular events. JAMA 2004;292:2343. 33. Lee DK, Marantz PR, Devereux RB, Kligfield P, Alderman MH. Left ventricular hypertrophy in black and white hypertensives. Standard ECG criteria overestimate racial differences in prevalence. JAMA 1992;267:3294. 34. Arnett DK, Rautaharju P, Crow R, et al, the ARIC Investigators. Blackwhite differences in ECG left ventricular mass and its association with blood pressure (the ARIC Study). Am J Cardiol 1994;74:247. 35. Rautaharju PM, Park LP, Gottdiener JS, et al. Race- and sex-specific ECG models for left ventricular mass in older populations. Factors influencing overestimation of left ventricular hypertrophy prevalence by ECG criteria in African-Americans. J Electrocardiol 2000;33:205. 36. Okin PM, Wright JT, Nieminen MS, et al, for the LIFE Study Investigators. Ethnic differences in electrocardiographic criteria for left ventricular hypertrophy: the LIFE Study. Am J Hypertens 2002;15: 663.
Journal of Electrocardiology 41 (2008) 342 – 350 www.jecgonline.com
Thresholds in the relationship between mortality and left ventricular hypertrophy defined by electrocardiography☆ Edward P. Havranek, MD, a,b, * Caroline D.B. Emsermann, MS, b Desiree N. Froshaug, BS, b Frederick A. Masoudi, MD MSPH, a,b Mori J. Krantz, MD, a,b Rebecca Hanratty, MD, a,b Raymond O. Estacio, MD, a,b L. Miriam Dickinson, PhD, b John F. Steiner, MD, MPH b b
a Department of Medicine, Denver Health Medical Center, Denver, Colorado, USA Colorado Health Outcomes Program, University of Colorado at Denver and Health Sciences Center, Denver, Colorado, USA Received 13 February 2007; accepted 15 November 2007
Abstract
Background: Electrocardiographic criteria for the diagnosis of left ventricular hypertrophy in current use were defined using autopsy results or echocardiography; criteria defined using mortality might be more clinically meaningful. Methods: Using data from Third National Health and Nutrition Examination Survey (NHANES III), we selected electrocardiographic measures that best differentiated those surviving at 5 years from those who did not. We identified voltage thresholds using regression techniques and then compared survival for subjects above and below the thresholds. Results: Cornell voltage, Cornell product, and Novacode estimate of left ventricular mass index were discriminative for mortality and had identifiable thresholds present in their relationships with mortality. Independent of systolic blood pressure, there were significant associations with 5-year mortality for Novacode index above threshold; hazard ratios were 1.58 for women and 1.27 for men, and for 5-year cardiovascular mortality were 1.78 for women and 2.34 for men. Conclusions: Electrocardiographic criteria for left ventricular hypertrophy validated against mortality might be clinically useful. © 2008 Elsevier Inc. All rights reserved.
Keywords:
Left ventricular hypertrophy; Electrocardiography; Mortality; Survival; NHANES
Introduction Left ventricular hypertrophy (LVH) is a significant risk factor for cardiovascular disease independent of blood pressure,1,2 and reductions in left ventricular mass are associated with lower cardiovascular event rates independent of reductions in blood pressure. In the Losartan Intervention for Endpoint reduction (LIFE) study, the risk of cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke was reduced by 22% for each one SD decrement in left ventricular mass index measured by echocardiography, independent of systolic blood pressure or treatment assigned.3 In the Heart Outcomes Prevention Evaluation (HOPE) trial, patients who had resolution of electrocardio☆ National Heart, Lung, and Blood Institute grant U01 HL079160 (Bethesda, Maryland, USA) and Health Resources and Services Administration Administrative Unit grant HP00054 5 D12 (Rockville, Maryland, USA). ⁎ Corresponding author. Denver Health Medical Center #0960, Denver, CO 80204-4507, USA. Tel.: +1 303 436 5499; fax: +1 303 436 7739. E-mail address: [email protected]
0022-0736/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.jelectrocard.2007.11.006
graphic LVH findings or who did not develop new LVH findings had significantly fewer cardiovascular deaths than did patients who did not have resolution or who developed new LVH.4 In the Appropriate Blood Pressure Control in Diabetes (ABCD) trial, the change in a QRS voltage measure of LVH predicted time to cardiovascular events, after controlling for change in blood pressure.5 These findings imply that hypertension treatment that leads to both regression of LVH and blood pressure reduction to goal may decrease cardiovascular event rates more than treatment to a blood pressure goal alone. To be clinically meaningful, a definition of LVH should be linked to a mortality reduction and make use of readily available technology. The most readily available technologies used to define LVH in research and in clinical practice are echocardiography and electrocardiography. Echocardiography has better sensitivity and specificity for detection of LVH than electrocardiography but is expensive and is technically inadequate in 10% to 20% of unselected subjects1 and in more than 30% in of older subjects.6 In contrast, electrocardiography is inexpensive,
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
is readily available, and can be performed in nearly all patients and interpreted with relatively little advanced skill. We hypothesized that electrocardiographic voltage measures of LVH treated as continuous variables have relationships with mortality that can be modeled to define a threshold above which mortality rises progressively, independent of blood pressure. Methods Data sources The data sources for the study were the Third National Health and Nutrition Examination Survey (NHANES III)7 database and the NHANES III linked mortality file.8 Third National Health and Nutrition Examination Survey was conducted in 81 counties in the United States between 1988 and 1994; subjects were selected to obtain a sample representative of the civilian, noninstitutionalized population of the United States greater than 2 months old. Electrocardiograms were obtained in NHANES III only for subjects older than 40 years. Twelve-lead tracings were recorded during examination at a mobile examination center by trained technicians using a Marquette MAC 12 system (GE Healthcare, UK). Electrocardiograms were transferred to the National Center for Health Statistics (Hyattsville, MD) for electronic analysis. Baseline blood pressure was measured as the average of the 3 recordings performed at the mobile examination center by a physician. Examiners used a mercury manometer and were instructed to record the blood pressure from the right arm with the subject seated, choosing the appropriate cuff size from among four available. The NHANES III linked mortality file contains the results of linking NHANES III subject identifiers with data in the National Death Index9 as of December 31, 2000. Date of death and cause of death available in the National Death Index are from death certificates. Definitions We chose to examine electrocardiographic measures of LVH that had been validated by echocardiography and that could be analyzed as continuous variables. Using these criteria we selected 6 measures. The Cornell voltage10 is the sum of the R wave dimension in lead aVL and the S wave dimension in lead V3; LVH is said to be present if this voltage is greater than 20 mm in women and 28 mm in men. The Cornell product11 is the product of the Cornell voltage and the QRS duration; LVH is said to be present if this product is greater than 2436 mm-ms (adjusted by the addition of 8 mm for female sex). The adjusted Cornell voltage is defined by the regression equations of Norman and Levy12: For males: ðCornell voltageþ0:0174 * ageþ0:1914 * BMI4:0Þ=2 For females: ðCornell voltageþ0:0387 * ageþ0:1778 * BMI4:9Þ=3 where BMI is the body mass index.
343 13
The Sokolow-Lyon voltage has been defined variously in the literature; we chose to define it as the sum of the greatest S wave dimension in leads V1 or V2 and the greatest R wave dimension in leads V5 or V6; LVH is said to be present if this voltage is greater than 35 mm. Finally, we used the Novacode estimates of left ventricular mass and left ventricular mass index calculated from the regression equations of Rautaharju et al.14 The variables and coefficients for these equations are available in the NHANES III electrocardiography data file documentation15 and in the Appendix. We also examined the independent effect of the ST-T changes associated with LVH (the so-called “strain pattern”) on outcomes, both separately and in combination with Cornell voltage and Cornell product. A strain pattern was judged to be present if there was horizontal or downsloping ST depression associated with T wave inversion in lead V5 and/or V6.16 We defined an analysis set by including those subjects with all the necessary data available to calculate the electrocardiographic measures and whose vital status was known. We excluded subjects with QRS duration greater than 0.12 seconds because validity in the presence of bundle branch block has not been established for all of the electrocardiographic measures. We defined 5-year mortality as the primary outcome of interest. All-cause mortality is less subject to ascertainment bias than is cardiovascular mortality, and the 5-year time frame has been commonly used in cardiovascular epidemiology as a compromise between the longer times necessary for effects of the exposure (in this case LVH) to accrue and the shorter times that exclude the accrual of competing causes of mortality. We defined cardiovascular death as those for which the principal cause of death was listed on the death certificate as International Classification of Diseases, Ninth Revision (ICD-9) codes 390.0-459.9 (for deaths before or during 1999) or ICD, 10th Revision (ICD-10) codes I10-15, I20-25, I50-51, I60-69, I70-73 (for deaths after 1999). In assessing the independent contribution of electrocardiographic LVH measures to mortality risk, we defined the relevant covariates as the Framingham risk factors17 of age, sex, smoking status, presence or absence of diabetes, systolic blood pressure, and total and high-density lipoprotein cholesterol. Diabetes was defined in NHANES III by self-identification. Statistical analysis We assessed for colinearity among the 6 measures using Pearson correlation coefficients. We selected the electrocardiographic measures best able to discriminate those at risk for mortality from those not at risk using receiver operating characteristic (ROC) curves. We divided the sample into deciles (with an equal number of subjects in each group), plotted 5-year all-cause mortality by decile, and calculated sensitivity and specificity for 5-year mortality by decile. We constructed a separate ROC curve for each of the 6 measures and calculated the area under the ROC curve. Measures were considered discriminative if the
344
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
lower bound of the 95% confidence intervals around the areas under the curve was greater than 0.50. For the discriminative measures, we sought thresholds above which mortality rose above a baseline level. Because left ventricular mass and left ventricular mass index vary systematically by sex, we sought separate thresholds for men and women. We selected at random two thirds of the analysis sample for a development sample. To enhance interpretability, we recoded the data for this set by 10-unit ranges for the measure, except for the smallest and largest of the ranges that included the remaining values on either end, for a total of 10 ranges. We applied piecewise regression to this set with 5-year mortality as the outcome, modeling the data as a spline function with 2 distinct slopes. We took as our thresholds the midpoint of the ranges for which risk was significantly greater than for values below that category. We tested the performance of the thresholds we found in 4 ways: First, we compared 5-year survival curves for male and female subjects with baseline voltage measures above and below the thresholds using the log-rank test. Second, we used Cox proportional hazards models to assess for an independent contribution of voltage, first controlling for systolic blood pressure alone and then controlling for all the Framingham risk factors, to the primary outcome of 5-year all-cause mortality. Third, we created the same Cox proportional hazards models for 3 secondary outcomes: 5-year cardiovascular mortality, 10-year all-cause mortality, and 10-year cardiovascular mortality. Finally, we repeated the Cox models in the one third of the sample that was not used to establish the thresholds. Results Of subjects in the NHANES III sample, 8561 had an electrocardiogram performed. We excluded from our
analysis sample the 641 subjects with QRS duration more than 0.12 second, 4 subjects because of incomplete electrocardiographic data, and 4 subjects because vital status was not known, leaving 7852 subjects with data available for analysis. In addition, 379 subjects were excluded from the adjusted analyses in the validation set because of missing Framingham risk factors. The characteristics of the study sample are shown in Table 1 and reflect a middle-aged and elderly population. The mean age of the sample was approximately 60 years and was slightly female predominant. Approximately 10% of subjects had diabetes, and approximately 40% had systolic hypertension. Correlation coefficients among the voltage measures are shown in Table 2; with the possible exception of the Cornell voltage and Cornell product, the measures are sufficiently different from one another to justify examination of each of them. Fig. 1 shows 5-year all-cause mortality by decile for each of the 6 electrocardiographic measures, and Table 3 shows the area under the ROC curve (C-statistic) for each of the 6 measures. Four measures, Cornell voltage, Cornell product, adjusted Cornell voltage, and Novacode estimate of left ventricular mass index, were discriminative for mortality, with C-statistics significantly greater than 0.5. Because we examined LVH as a univariate predictor of mortality, the C-statistics are much lower than that typically described for multivariable models. By piecewise regression analysis, there were statistically identifiable thresholds in the mortality relationship for 3 of the discriminative measures: Cornell voltage, Cornell product, and Novacode left ventricular mass index; these thresholds are shown in Table 3. In men, 16.7% of the sample had a Novacode left ventricular mass index above threshold, 16.1% had a Cornell voltage above threshold, and 13.3% had Cornell product above threshold. Comparable figures for women were 19.3%, 9.2%, and 13.2%, respectively.
Table 1 Characteristics of study population Variable
Total
Males
Females
Age (y, mean [SD]) Sex (% female) Current smoker at baseline (%) Total cholesterol (mmol/L, mean [SD]) HDL cholesterol (mmol/L, mean [SD]) Diabetes (%) Systolic BP (mm Hg) Cornell voltage (mm, mean [SD]; median, IQR) Cornell product, mm-ms, (mean [SD]; median, IQR) Adjusted Cornell voltage (mean [SD]; median [IQR]) Sokolow-Lyons voltage (mm, mean [SD]; median [IQR]) Novacode mass (g, mean [SD]; median [IQR]) Novacode mass index (g/m2, mean [SD]; median [IQR])
59.9 (13.4) 53 23 5.64 (1.14) 1.32 (0.42) 11 132.9 (19.9) 13.1 (5.5); 12.7 (9.20-16.5)
60 (13.4) 27 5.51 (1.10) 1.19 (0.38) 11 134.0 (18.5) 14.0 (5.7); 14.0 (10.3-17.9)
59.7 (13.5) 18 5.78 (1.16) 1.42 (0.42) 11 132.4 (21.0) 12.0 (5.0); 11.6 (8.38-15.1)
1272 (580); 1205 (858-1611)
1428 (610); 1368 (999-1788)
1132 (512); 1076 (761-1432)
1.5 (0.6); 1.5 (1.08-1.90)
1.8 (0.6); 1.8 (1.44-2.19)
1.2 (0.5); 1.2 (0.88-1.50)
25.9 (7.8); 25.0 (20.6-30.2)
27 (8.3); 26.3 (21.5-32.1)
24.8 (7.0); 23.8 (20.0-28.7)
154 (29.4); 153 (133-174)
172 (25.5); 171 (156-188)
138 (22.6); 136 (123-152)
105 (21.7); 103 (90-118)
112 (20.5); 110 (99-123)
98.4 (20.8); 96.0 (85-110)
HDL indicates high-density lipoprotein; BP, blood pressure; IQR, interquartile range.
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
345
Table 2 Pearson correlation coefficient matrix
Cornell voltage Cornell product Adjusted Cornell Sokolow-Lyons Novacode LVM Novacode LVMI
Cornell voltage
Cornell product
Adjusted Cornell
Sokolow-Lyons
Novacode LVM
Novacode LVMI
1.0 0.942 0.628 0.397 0.566 0.465
1.0 0.621 0.357 0.631 0.585
1.0 0.093 0.855 0.393
1.0 0.095 0.401
1.0 0.471
1.0
LVM indicates left ventricular mass; LVMI, left ventricular mass index.
As described in the Methods section, we tested the thresholds in 4 ways. First, we compared 5-year survival for women and for men by voltage above and below the thresholds. As shown in Figs. 2, 3, and 4, there are statistically significant differences for all 3 sets of thresholds. Second, we assessed for an independent contribution of the voltages to 5-year mortality by adjusting first for systolic blood pressure alone and then for all the Framingham risk factors. As shown in Table 4, statistically significant relationships for men are present with adjustment for all the Framingham risk factors but not for blood pressure alone. For women, there are statistically significant relationships
adjusted for blood pressure but not for all the Framingham risk factors. Third, we created the same Cox proportional hazards models for the 3 secondary outcome measures of 5-year cardiovascular mortality, 10-year all-cause mortality, and 10-year cardiovascular mortality. Adjusted relationships were stronger for cardiovascular mortality than for allcause mortality. Fig. 5 shows the results for the adjustment for systolic blood pressure alone. The Novacode left ventricular mass index predicted mortality independent of blood pressure for all 3 secondary outcome measures; results with the other 2 measures were less consistent. With adjustment for all Framingham risk factors in men, the
Fig. 1. Five-year all-cause mortality by decile for each of the voltage criteria tested.
346
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
Table 3 C-statistics and sex-specific thresholds for each of the voltage criteria Voltage measure
C-statistic (95% confidence interval)
Threshold—females
Threshold—males
Cornell voltage Cornell product Adjusted Cornell voltage Sokolow-Lyons voltage Novacode LVM Novacode LVMI
0.55 (0.53-0.57) 0.55 (0.56-0.57) 0.56 (0.54-0.59) 0.52 (0.49-0.54) 0.50 (0.48-0.52) 0.59 (0.57-0.61)
19 mm 1700 mm-ms Not applicable ⁎ Not applicable Not applicable 115 g/m2
20 mm 2150 mm-ms Not applicable Not applicable Not applicable 130 g/m2
⁎ Piecewise regression demonstrated there was no threshold for the adjusted Cornell voltage.
thresholds for all 3 measures predicted mortality independently for all 3 secondary outcomes. For women, the thresholds predicted mortality independent of all Framingham risk factors only for Novacode left ventricular mass index and 10-year cardiovascular mortality. Finally, we tested these thresholds in the one third of the sample not used to develop the thresholds. As shown in Table 5, hazard ratios were consistent with those found in the two thirds of the sample used to develop the thresholds for Novacode left ventricular mass index but not for the Cornell voltage and Cornell product. Results were similar for the secondary outcomes in the validation set. We performed additional analysis to assess whether the ST-T changes accounted for in the Novacode left ventricular mass index were responsible for the superior performance of this measure. First, we assessed the independent contribution of the strain pattern alone. In males in the development set, the hazard ratio associated with strain for 5-year all-cause
Fig. 2. A, Kaplan-Meier curves comparing survival for men by Novacode left ventricular mass index above and below threshold. B, Kaplan-Meier curves comparing survival for women by Novacode left ventricular mass index above and below threshold.
mortality was 5.72 (3.67-8.92) unadjusted, 4.56 (2.88-7.23) adjusted for systolic blood pressure, and 2.56 (1.59-4.13) adjusted for all Framingham risk factors; comparable hazard ratios for women were 4.84 (2.28-10.29), 2.44 (1.13-5.29), and 2.11 (0.93-4.81), respectively. Results in the validation set confirmed these findings. The value of this strong association between the strain pattern and subsequent mortality was tempered by its relatively low prevalence. A strain pattern was found on the electrocardiograms of only 5.0% of the 760 subjects who had died at 5 years, whereas 18.5% of subjects with Cornell voltage above threshold, 18.2% of subjects with Cornell product above threshold, and 26.8% of subjects with Novacode left ventricular mass index above threshold had died at 5 years. We therefore tested association between presence of a strain pattern in combination with Cornell voltage and Cornell product. In males in the development set, the hazard ratio associated with strain combined with Cornell voltage for 5-year all-cause mortality was 1.68 (1.30-2.18) unadjusted, 1.42 (1.09-1.86) adjusted for systolic blood pressure,
Fig. 3. A, Kaplan-Meier curves comparing survival for men by Cornell voltage above and below threshold. B, Kaplan-Meier curves comparing survival for women by Cornell voltage above and below threshold.
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
347
These results suggest that electrocardiography might be used to establish regression of left ventricular hypertrophy in the management of hypertension, which in turn might be expected to reduce mortality. Even small increases in voltages above the thresholds appear to be associated with significant increases in mortality. Comparison with prior literature
Fig. 4. A, Kaplan-Meier curves comparing survival for men by Cornell product above and below threshold. B, Kaplan-Meier curves comparing survival for women by Cornell product above and below threshold.
and 1.47 (1.12-1.94) adjusted for all Framingham risk factors; comparable hazard ratios for women were 2.85 (2.04-3.97), 1.76 (1.23-2.52), and 1.44 (0.99-2.10), respectively. These results were marginally better than the Novacode measure in men and marginally worse than the Novacode measure in women and were nearly identical for the Cornell product. Results in the validation set confirmed these findings. Of the 760 subjects who died, 21.4% had strain or Cornell voltage criteria present and 21.2% had strain or Cornell product present. Discussion Our study confirmed the relationship between greater electrocardiographic voltage measures of LVH and higher mortality and identified the Cornell voltage, Cornell product, and Novacode estimate of left ventricular mass index as having discriminative capacity for mortality and having demonstrable thresholds above which mortality rises progressively. The Novacode left ventricular mass index performed slightly better in multivariable models. The Cornell voltage and Cornell product also demonstrated significant threshold relationships, but the thresholds showed fewer significant relationships with mortality after adjustment for other risk factors. Novacode left ventricular mass index above threshold was present in 26.8% of subjects who had died at 5 years; for Cornell voltage and Cornell product these figures were 18.5% and 18.2%, respectively. There were fewer statistically significant relationships in a validation sample comprising one third of the full analytic sample probably because of the smaller number of observations in the validation sample.
Electrocardiographic measures of LVH have been associated with elevated risk of mortality in a number of prior studies, though few have treated these measures as continuous variable. Prineas and colleagues18 showed the Sokolow-Lyons voltage, Cornell voltage, Cornell product, and Novacode score used as continuous variables were better predictors of mortality in patients in the Multiple Risk Factor Intervention Trial than were the same parameters dichotomized as the presence or absence of LVH. The electrocardiographic measure that performed best in our study, the Novacode estimate of left ventricular mass index, is probably the least studied of the measures we investigated. In one study,19 this score had lower sensitivity and specificity for the detection of echocardiographic LVH than the Sokolow-Lyons voltage, the Cornell voltage, and the Cornell product. In a study of subjects in the Cardiovascular Health Study,6 the Novacode equations consistently overestimated left ventricular mass. The superior sensitivity and specificity for detection of mortality despite its limitations for detection of echocardiographic hypertrophy may be a result of the unique use of T-wave inversion in the Novacode score. Inversion of the T wave is a feature of the “strain” pattern that is an independent predictor of poor prognosis.20 This explanation is supported by the strong association between the presence of a strain pattern and subsequent mortality in our data set and by combination of the strain pattern with either Cornell voltage or Cornell product made the performance of these 2 measures comparable to that of the Novacode index. Further Table 4 Associations between 5-year total mortality and electrocardiographic indices above threshold for the development set Males (309 deaths)
Unadjusted Adjusted for systolic BP Adjusted for all Framingham risk factors ⁎
Cornell voltage
Cornell product
Novacode LVMI
HR
HR
HR
95% CI
95% CI
95% CI
1.42 1.08-1.87 1.38 1.03-1.86 1.52 1.16-1.98 1.19 0.89-1.58 1.16 0.85-1.57 1.27 0.99-1.68 1.34 1.01-1.79 1.34 0.98-1.82 1.40 1.05-1.85
Females (201 deaths) Unadjusted 2.66 1.89-3.75 2.33 1.70-3.20 2.33 1.74-3.12 Adjusted for 1.65 1.15-2.24 1.60 1.15-2.24 1.58 1.16-2.16 systolic BP Adjusted for 1.34 0.91-1.98 1.41 0.99-2.01 1.12 0.81-1.54 all Framingham risk factors HR indicates hazard ratio; CI, confidence interval. ⁎ Age, smoking status, diabetes, systolic blood pressure, total cholesterol, HDL cholesterol.
348
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
Fig. 5. Hazard ratios with 95% confidence intervals for the secondary outcome measures, adjusted for systolic blood pressure, for men (top panel) and for women (bottom panel).
development and validation of such combinations are worthy of future study. The thresholds we identified for the Novacode left ventricular mass index are strikingly comparable to the upper limits of normal for echocardiographic left ventricular mass index determined by Devereux and colleagues21 (132136 g/m2 for men and 109-112 g/m2 for women). In an article using data from the Framingham study and the same echocardiographic conventions as Devereux et al, the thresholds identified were similar for men (131 g/m2) and slightly lower for women (100 g/m2).22 The hazard ratios for mortality in the current study are similar to the results of studies based on echocardiography for men but not for women. In a report from the Framingham study, Levy et al1 found relative risk of 1.49 in men for allcause mortality adjusted for age and risk factors and 1.73 for cardiovascular mortality; for women, the comparable figures were 2.01 and 2.12, respectively. Liao and colleagues23 studied the relationship between echocardiographically
determined left ventricular mass and subsequent mortality and, after adjusting for age and hypertension, also found a stronger relationship between LVH and mortality in women compared to men. In the current study, the opposite is the case, with relationships stronger in men than in women. This suggests that further refinement of the Novacode equations for women is possible and might further improve the discriminative capacity of the thresholds. Although the Novacode left ventricular mass index threshold was more clearly predictive of mortality than the Cornell voltage or the Cornell product thresholds, both of these latter measures were discriminative for mortality and had identifiable thresholds. For men, we found an increase in mortality at lower levels than the established criteria for diagnosis of LVH for both measures (20 vs 28 mm for the Cornell voltage and 2150 vs 2436 mm-ms for the Cornell product). For women, we found thresholds at comparable levels (19 vs 20 mm for the Cornell voltage and 1700 vs approximately 1740 mm-ms for the Cornell product). The
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350 Table 5 Associations between 5-year total mortality and electrocardiographic indices above threshold for the validation set Males (153 deaths)
Unadjusted Adjusted for systolic BP Adjusted for all Framingham risk factors ⁎
Cornell voltage
Cornell product
Novacode LVMI
HR
HR
95% CI
HR
1.02 0.64-1.62 1.08 0.82 0.51-1.31 0.88
0.65-1.80 0.53-1.46
1.43 0.97-2.11 1.14 0.76-1.70
0.86 0.53-1.40 0.91
0.53-1.55
1.29 0.86-1.92
95% CI
Females (118 deaths) Unadjusted 1.83 1.04-3.21 1.45 Adjusted for 1.47 0.82-2.64 1.18 systolic BP Adjusted for 1.03 0.53-1.97 0.84 all Framingham risk factors
95% CI
349
attenuated when the higher prevalence of obesity in African Americans is accounted for.35,36 The Novacode estimate of left ventricular mass index not only accounts for differences in body size but also uses separate regression equations for black and white women. Other potential ethnic differences, particularly for Latinos, have generally not been explored in the literature. The relationship among race and ethnicity, electrocardiographic measures of LVH, and mortality is deserving of future research. Conclusions and implications
0.87-2.42 0.70-2.00
2.19 1.44-3.32 1.84 1.17-2.88
0.48-1.49
1.21 0.76-1.92
⁎ Age, smoking status, diabetes, systolic blood pressure, total cholesterol, HDL cholesterol.
Electrocardiographic definitions of LVH based on relationships with mortality are feasible. Such definitions might be clinically useful in the routine management of patients with hypertension as goals for therapy in addition to lowering blood pressure below goal; independent prospective validation is required. Appendix Men:
reasons for this sex difference in the relationship between electrocardiographic voltage and outcome are uncertain. One possibility is that the partition values for Cornell voltage and Cornell product were set too high and that partitions with lower sensitivities would be more clinically useful. The Cornell voltage and the Cornell product are simple to calculate, might therefore be more useful in routine clinical practice, and are worthy of further study. The poor performance of the Sokolow-Lyons voltage is in part because of its low sensitivity and specificity for the detection of LVH,24-28 but other factors may be at play. A study by Velury and Spodick29 suggested that low R voltage (b10 mm) in leads V5 and V6 has a high sensitivity for the diagnosis of LVH. This may explain the trend toward higher mortality in patients with low Sokolow-Lyons voltage seen in Fig. 1 because low Sokolow-Lyon voltage might also be a marker of LVH. Study limitations The greatest limitation to the current study is that it relies on data taken at a single point in time. We are thus unable to assess whether decrements in the electrocardiographically estimated left ventricular mass index are associated with diminished risk of mortality. Data from other studies,30-32 however, suggest an association between decrements in electrocardiographic voltage and diminished mortality. Information about potentially significant cofactors, such as treatment during follow-up, was not available. Another limitation is that we relied on cardiovascular causes of death determined from death certificates, which may be unreliable. However, we found significant relationships with all-cause mortality, an outcome less subject to ascertainment bias. We chose not to consider race or ethnicity as possible covariates. Although the prevalence of LVH assessed by Sokolow-Lyon or Cornell voltage criteria is greater in African Americans, there are no racial differences when assessed by echocardiography,33,34 and the diminished specificity of the electrocardiographic criteria are greatly
Variable R in amplitude in V5 Q or S amplitude in V1 Q or S amplitude in III Negative T amplitude in V6 Positive T amplitude in aVR QRS duration Intercept
Coefficient LV mass 0.0217 0.0338 0.0600 0.3158 −0.2958 1.8204 −58.5098
LV mass index 0.0100 0.0203 0.0287 0.1819 −0.1482 1.0485 −36.4290
Coefficient LV mass 0.0320 0.0233 0.0693 −0.1545 0.1122 – −0.1236 134.7722
LV mass index – 0.0178 0.0528 −0.1128 0.1075 0.1701 −0.0939 88.4357
White women:
Variable R in amplitude in aVL R amplitude in V5 Q or S amplitude in V5 Q or S amplitude in I Positive T amplitude in V1 Negative T amplitude in aVF Positive T amplitude in V6 Intercept Black women:
Variable R in amplitude in aVL R amplitude in I R amplitude V6 + S amplitude V2 R amplitude in V1 R amplitude in V2 Larger of Q or S amplitude in V6 Negative T amplitude in aVL Negative T amplitude in I QRS duration Intercept
Coefficient LV mass – 0.0498 0.0235 −0.0507 – −0.0980 – 0.5225 1.8478 −90.7136
LV mass index 0.0216 – 0.0184 – −0.0143 −0.0693 0.199 – 0.7460 −22.3064
References 1. Levy D, Garrison RJ, Savage DD, Kannel WB, Casatelli WP. Prognostic implications of echocardiographically determined left ventricular mass in the Framingham Heart Study. N Engl J Med 1990;322:1561.
350
E.P. Havranek et al. / Journal of Electrocardiology 41 (2008) 342–350
2. Brown DW, Giles WH, Croft JB. Left ventricular hypertrophy as a predictor of coronary heart disease mortality and the effect of hypertension. Am Heart J 2000;140:848. 3. Devereux RB, Wachtell K, Gerdts E, et al. Prognostic significance of left ventricular mass change during treatment of hypertension. JAMA 2004;292:2350. 4. Matthew J, Sleight P, Lonn E, et al, for the Heart Outcomes Prevention Evaluation (HOPE) Investigators. Reduction of cardiovascular risk by regression of electrocardiographic markers of left ventricular hypertrophy by the angiotensin-converting enzyme inhibitor ramipril. Circulation 2001;104:1615. 5. Havranek EP, Esler A, Mehler PS, Estacio RO, Schrier RW. Differential effects of anti-hypertensive agents on ECG voltage: results from the Appropriate Blood Pressure Control in Diabetes (ABCD) Trial. Am Heart J 2003;145:993. 6. Rautaharju PM, Manolio TA, Siscovick D, et al, for the Cardiovascular Health Study Collaborative Research Group. Utility of new electrocardiographic models for left ventricular mass in older adults. Hypertension 1996;28:8. 7. http://www.cdc.gov/nchs/nhanes.htm. Accessed June 1, 2007. 8. http://www.cdc.gov/nchs/r&d/nchs_datalinkage/nhanes_data_linkage_ activities.htm. Accessed June 1, 2007. 9. http://www.cdc.gov/nchs/ndi.htm. Accessed June 1, 2007. 10. Casale PN, Devereux RB, Kligfield P, et al. Electrocardiographic detection of left ventricular hypertrophy: development and prospective validation of improved criteria. J Am Coll Cardiol 1985;6:572. 11. Molloy TJ, Okin PM, Devereux RB, Kligfield P. ECG detection of echocardiographic left ventricular hypertrophy by the simple QRS voltage-duration product. J Am Coll Cardiol 1992;20:1180. 12. Norman JE, Levy D. Improved electrocardiographic detection of echocardiographic left ventricular hypertrophy: results of a correlated data base approach. J Am Coll Cardiol 1995;26:1022. 13. Sokolow M, Lyon TP. The ventricular complex in left ventricular hypertrophy as obtained by unipolar precordial and limb leads. Am Heart J 1949;37:161. 14. Rautaharju PM, Lacroix AZ, Savage DD, et al. Electrocardiographic estimate of left ventricular mass versus radiographic cardiac size and the risk of cardiovascular disease mortality in the epidemiologic Follow-Up Study of the First National Health and Nutrition Examination Survey. Am J Cardiol 1988;62:59. 15. Available at http://www.cdc.gov/nchs/about/major/nhanes/nh3data. htm#Data%20Files%202a. Accessed June 1, 2007. 16. Okin PM, Devereux RB, Nieminen MS, et al, for the LIFE Study Investigators. Relationship of the electrocardiographic strain pattern to left ventricular structure and function in hypertensive patients: The LIFE Study. J Am Coll Cardiol 2001;38:514. 17. Anderson KM, Wilson PWF, Odell PM, Kannel WB. An updated coronary risk profile. A statement for health professionals. Circulation 1991;83:357. 18. Prineas RJ, Rautaharju PM, Grandits G, Crow R, for the MRFIT Research Group. Independent risk for cardiovascular disease predicted by modified continuous score ECG criteria for 6-year incidence and regression of left ventricular hypertrophy among clinically disease free men: 16-year follow-up for the Multiple Risk Factor Intervention Trial. J Electrocardiol 2001;34:91. 19. Okin PM, Roman MJ, Devereux RB, Kligfield P. Electrocardiographic identification of left ventricular hypertrophy: test performance in relation to definition of hypertrophy and presence of obesity. J Am Coll Cardiol 1996;27:124.
20. Verdecchia P, Schillaci G, Borgioni C, et al. Prognostic value of a new electrocardiographic method for diagnosis of left ventricular hypertrophy. J Am Coll Cardiol 1998;31:383. 21. Devereux RB, Lutas EM, Casale PN, et al. Standardization of M-mode echocardiographic left ventricular anatomic measurements. J Am Coll Cardiol 1984:1222. 22. Levy D, Savage DD, Garrison RJ, Anderson KM, Kannel WB, Castelli WP. Echocardiographic criteria for left ventricular hypertrophy: the Framingham Heart Study. Am J Cardiol 1987;59:956. 23. Liao Y, Cooper RS, Mensah GA, McGee DL. Left ventricular hypertrophy has a greater impact on survival in women than in men. Circulation 1995;92:805. 24. Casale PN, Devereux RB, Kligfield P, et al. ECG detection of left ventricular hypertrophy: development and prospective validation of improved criteria. J Am Coll Cardiol 1985;6:572. 25. Casale PN, Devereux RB, Alonso DR, Campo E, Kligfield P. Improved sex-specific criteria of left ventricular hypertrophy for clinical and computer interpretation of electrocardiograms: validation with autopsy findings. Circulation 1987;75:565. 26. Vijan SG, Manning G, Millar-Craig MW. How reliable is the electrocardiogram in detecting left ventricular hypertrophy in hypertension? Postgrad Med J 1991;67:646. 27. Abergel E, Tase M, Menard J, Chatellier G. Influence of obesity on the diagnostic value of ECG criteria for detecting left ventricular hypertrophy. Am J Cardiol 1996;77:739. 28. Okin PM, Jern S, Devereux RB, Kjeldsen SE, Dahlof B, for the LIFE Study Group. Effect of obesity on electrocardiographic left ventricular hypertrophy in hypertensive patients: The Losartan Intervention For Endpoint (LIFE) Reduction in Hypertension Study. Hypertension 2000;35:13. 29. Velury S, Spodick DH. Increased left ventricular mass in patients with lateral low-voltage electrocardiograms. Am J Cardiol 1992;69:707. 30. Levy D, Salomon M, D'Agostino RB, Belanger AJ, Kannel WB. Prognostic implications of baseline ECG features and their serial changes in subjects with left ventricular hypertrophy. Circulation 1994;90:1786. 31. Fagard RH, Staessen JA, Thijs L, et al, for the Systolic Hypertension in Europe (Syst-Eur) Trial Investigators. Prognostic significance of electrocardiographic voltages and their serial changes in elderly with systolic hypertension. Hypertension 2004;44:459. 32. Okin PM, Devereux RB, Jern S, et al, for the LIFE Study Investigators. Regression of electrocardiographic left ventricular hypertrophy during antihypertensive treatment and the prediction of major cardiovascular events. JAMA 2004;292:2343. 33. Lee DK, Marantz PR, Devereux RB, Kligfield P, Alderman MH. Left ventricular hypertrophy in black and white hypertensives. Standard ECG criteria overestimate racial differences in prevalence. JAMA 1992;267:3294. 34. Arnett DK, Rautaharju P, Crow R, et al, the ARIC Investigators. Blackwhite differences in ECG left ventricular mass and its association with blood pressure (the ARIC Study). Am J Cardiol 1994;74:247. 35. Rautaharju PM, Park LP, Gottdiener JS, et al. Race- and sex-specific ECG models for left ventricular mass in older populations. Factors influencing overestimation of left ventricular hypertrophy prevalence by ECG criteria in African-Americans. J Electrocardiol 2000;33:205. 36. Okin PM, Wright JT, Nieminen MS, et al, for the LIFE Study Investigators. Ethnic differences in electrocardiographic criteria for left ventricular hypertrophy: the LIFE Study. Am J Hypertens 2002;15: 663.